Print events from matplotlib event handlers get swallowed

[fperez]

Consider the following code in the notebook:

%pylab notebook

fig, ax = plt.subplots()
ax.plot(np.random.rand(10))

def onclick(event):
    print ('button=%d, x=%d, y=%d, xdata=%f, ydata=%f'%(
        event.button, event.x, event.y, event.xdata, event.ydata))

cid = fig.canvas.mpl_connect('button_press_event', onclick)

these events never print anywhere.

The code does run, as can be seen by this code from @tacaswell:

%pylab notebook

fig, ax = plt.subplots()
ax.plot(np.random.rand(10))
collector = []
def onclick(event):
    collector.append('button=%d, x=%d, y=%d, xdata=%f, ydata=%f'%(
        event.button, event.x, event.y, event.xdata, event.ydata))

cid = fig.canvas.mpl_connect('button_press_event', onclick)

After clicking a couple of times:

In [2]: collector
['button=1, x=201, y=275, xdata=2.196853, ydata=0.615071',
 'button=1, x=305, y=360, xdata=4.100549, ydata=0.770019',
 'button=1, x=421, y=223, xdata=6.205388, ydata=0.520279']

Somehow we're swallowing that stdout completely, I'm not 100% sure why right now. But to the user, it feels very puzzling. Given the equivalent code works just fine at the terminal, we should try to make it work without jumping through additional hoops in the notebook.

[takluyver]

I think it's another case of "where should the output go". The kernel is probably sending the output, but the notebook frontend doesn't know which cell to associate it with, so it discards it. I seem to recall @jdfreder spent some effort working round this for widgets. It would be good to have a more general solution, but I'm not sure what shape that would take.

[jdfreder]

Yeah, this is exactly the kind of thing I had to tackle with the widgets. The notebook isn't smart enough to know what cell triggered the click event, so you need to tell it before sending the message to the backend.

Thanks for pinging me here @takluyver , this is one of the things I'm trying to reimagine with the output area work I'm doing.

[minrk]

Threaded output is associated with the most recent execution (by the kernel). With a GUI backend, I see the output in the cell (whichever is the most recently executed). With the nbagg backend, the most recent request is likely to be a comm message, and the nbagg backend is presumably dropping those messages. The nbagg backend could hook up the cell output callback for output messages it receives for its comms.

[fperez]

Mmmh... @ellisonbg, this is something that perhaps Ryan could work on as part of the MPL work.. . He might be able to make an improvement to the actual nbagg backend in MPL itself, working with @tacaswell (perhaps with some guidance from @minrk). What do you think?

[tacaswell]

@blink1073 is also working on a widget based version of nbagg

[blink1073]

Yes, see #4582 for where I left off. matplotlib/matplotlib#4582

[blink1073]

The issue I ran in to originally that the div created by mpl.show was being changed from under its nose during interact.

[blink1073]

When the following cell is executed, the comm_manager creates an nbagg connection, which creates a div. Might there be a way to add the nbagg div to the list of widgets in interactive?

%matplotlib notebook
import matplotlib.pyplot as plt
from IPython.html.widgets import interactive

fig, ax = plt.subplots()
ax.plot(range(5))


vline = ax.axvline(1, color='k')
hline = ax.axhline(0.5, color='k')

def set_cursor(x, y):
    vline.set_xdata((x, x))
    hline.set_ydata((y, y))
    ax.figure.canvas.draw_idle()

interactive(set_cursor, x=ax.get_xlim(), y=ax.get_ylim())

[fperez]

@jdfreder, perhaps you have some insights that can help us here?

[jdfreder]

@blink1073 if you want the div to be treated as something that can be used by interact, you'll need to write a custom widget view and python widget. The value traitlet of the widget will be what interact reads/writes.

[jdfreder]

If instead you want something that doesn't display in the widget area, you may be able to fool it by having a Python class that inherits from HasTraits, having a traitlet called value. Interact would read/write that value like it was a widget.

[tacaswell]

Not sure this should be closed. We also have an open mpl issue related to exceptions from event callbacks also being swallowed matplotlib/matplotlib#5111